THICKNESS TAILORING OF VARIABLE STIFFNESS PANELS FOR MAXIMUM BUCKLING LOAD

Size: px

Start display at page:

Download "THICKNESS TAILORING OF VARIABLE STIFFNESS PANELS FOR MAXIMUM BUCKLING LOAD"

Julius Miles
5 years ago
Views:

1 THICKNESS TAILORING OF VARIABLE STIFFNESS PANELS FOR MAXIMUM BUCKLING LOAD Samuel T. IJsselmuiden, Mostafa M. Abdalla, Zafer Gürdal Aerospace Structures Chair, Delft University of Technology Kluyverweg 1, 2629 HS Delft, The Netherlands SUMMARY By steering the composite fibres in curvilinear paths, spatial variation of stiffness can be induced resulting in beneficial load and stiffness distribution patterns. One especially relevant area in which fibre steering has proved its effectiveness is in improving buckling loads of composite panels. In addition, stiffness tailoring has also proved to be an efficient means of increasing load carrying capacity of plates. In this paper a separable approximation scheme is introduced which allows buckling loads to be maximised by simultaneously tailoring both stiffness and thickness. Local stiffness properties are expressed in terms of lamination parameters and thickness. Preliminary numerical results indicate that performance improvements in excess of 440 % are possible using this design scheme. Keywords: Composite Design, Variable Stiffness, Lamination Parameters, Buckling. INTRODUCTION Advanced fibre placement machines have made it possible to manufacture complex composite parts with consistent quality. The built in steering and cut/restart capabilities can be used to fully exploit the anisotropic nature of composite materials. Past research has shown that panel buckling loads improve significantly when stiffness properties are varied over the structural domain [10]. Biggers and Srinivasan [2] demonstrated that distributing stiff fibres to a plates edge results in higher buckling loads. Later Joshi and Biggers [7] presented a method to determine optimal thickness distributions for isotropic and anisotropic plates for maximum buckling loads. Kassapoglou et al. [8] presented results for which panel buckling loads improved by including a central patch, locally improving the bending stiffness. Buckling load improvements in the afore mentioned publications result from two fundamentally different mechanisms. Biggers and Joshi demonstrated that improvements were found to be primarily attributed to in-plane load redistribution. Results shown by Kassapoglou indicate that buckling loads increase due to local improvements in bending

2 stiffness. In previous work, the authors demonstrated that buckling loads improve primarily due to load redistribution [5] for constant thickness, variable stiffness panels. However, bending stiffness and therefore buckling, is highly dependent on thickness. Hence it is interesting to investigate the influence of thickness variations on the buckling response of variable stiffness panels as well to assess the mechanisms responsible for the improved buckling loads. To achieve this a conservative approximation scheme is developed which allows the buckling load to be optimised both in terms of thickness and lamination parameters. A finite element scheme to asses buckling loads is presented briefly in the next section followed by the definition of lamination parameters and their feasible region. Subsequent sections focus on the development of a conservative approximation scheme, initial optimisation results and several concluding remarks. BUCKLING ANALYSIS The buckling load is determined using a finite element discretization of the buckling analysis through the following eigenvalue problem, ( K m λk g) a = 0, (1) where K m is the global material stiffness matrix, K g is the global geometric stiffness matrix, a is the mode shape comprising of deformation degrees of freedom, and λ is load multiplier or buckling factor. The mode shapes are normalised such that, a T K m a = 1. (2) The geometric stiffness matrix is constructed through an assembly of element geometric matrices. The stiffness matrix of each element takes the form, K g e = n x K x n y K y n xy K xy, (3) where n e = (n x, n y, n xy ) T is the vector of in-plane stress resultants averaged over the element, and K x, K y and K xy are constant matrices that depend only on element geometry. The averaged in-plane stress resultants can be expressed as, n e = A e e e, (4) where A is the in-plane stiffness matrix and e is the average strain vector given by, e e = B e u e, (5) where u is the vector of displacements, B is the average element strain displacement matrix, and u e is the vector of the degrees of freedom associated with nodes connected to the e-th element. The displacements can be found from the solution of the equilibrium equations, where f is the vector of applied loads. K m u = f. (6)

3 LAMINATION PARAMETERS Stiffness properties of any given laminate can be uniquely defined by a set of twelve continuous parameters known as lamination parameters. For symmetric laminates eight parameters are sufficient, imposing an additional balanced condition reduces the number of parameters to four. Initially introduced by Tsai et al. [14, 15], the in-plane and out of plane lamination parameters are defined as: (V 1, V 2, V 3, V 4 ) = (W 1, W 2, W 3, W 4 ) = 12 1/2 1/2 1/2 (cos 2θ( z), sin 2θ( z), cos 4θ( z), sin 4θ( z))d z (7) 1/2 z 2 (cos 2θ( z), sin 2θ( z), cos 4θ( z), sin 4θ( z))d z (8) where z = z/h is the normalised through-the-thickness coordinate of the layers [6], h is the total thickness of the laminate, and θ( z) is the fibre angle at z. The in-plane laminate stiffness matrix A and the bending stiffness matrix D can subsequently be defined as a linear function of the lamination parameters: A = h(γ 0 + V 1 Γ 1 + V 2 Γ 2 + V 3 Γ 3 + V 4 Γ 4 ), (9) D = h3 12 (Γ 0 + W 1 Γ 1 + W 2 Γ 2 + W 3 Γ 3 + W 4 Γ 4 ) (10) where Γ i (i = 1,..., 4) are matrices in terms of laminate invariants, defined in appendix. Lamination parameters are not independent since the trigonometric functions used in Equation (7) are related. The feasible region is well defined [4] when solving design problems which are dependent only on in-plane or out of plane lamination parameters. However, the buckling load is not only a function of the bending stiffness, D, but also of the in-plane stiffness A [10]. Thus a feasible region of the combined in-plane and out of plane lamination parameters is required. Currently no analytical expression for the combined feasible domain is known, therefore it is approximated using a set of linear constraints obtained from successive convex hull approximations of the design space as was done by Setoodeh et al [12]. OPTIMISATION FORMULATION Successive approximation techniques are popular for structural optimisation as they increase computational efficiency by reducing the number of finite element analyses necessary to find the optimum. The idea is to create a separable approximation of the objective function and constraints such that each term in the approximation depends only on the design variables associated with one node (or element). The optimisation is carried out on the approximation after which the design is updated. This procedure is repeated until a converged design is found. The objective is to design the thickness and lamination parameter distribution such that the buckling load is maximised subject to an upper bound on the volume, v 0. The

4 problem is formulated in terms of the inverse buckling factor, r = 1/λ, which is a measure of structural compliance, as, min r s.t v v 0 (11) V i,w i,h i where V i, W i are the in-plane and out of plane lamination parameters, and h i is the thickness, of each design region, i. The lamination parameters are also bound by their feasible region. The total volume, v, is simply the sum of the product of thickness and area, A i, of each design region. The design regions can be defined at node level, element level, or span over multiple elements. In the current work, design variables are defined at nodes and element properties are computed via reciprocal interpolation [1]. The formulation presented in (11) assumes that only one eigenvalue is present during optimisation, however for multimodal problems all critical buckling modes must be incorporated in the design. Multimodal optimisation can be achieved using a method similar to the bound formulation proposed by Olhoff [9] by introducing an independent parameter, β, such that the minimisation problem becomes, min β s.t. β r n (12) v v 0 where r n (for n = 1, 2,... N) are the inverse buckling factors corresponding to the N critical buckling modes. The problem can subsequently be solved using the dualmethod [3, 13], which leads to the following two conditions, L C = min N µ n r n + ν n=1 ( ) v 1 v 0 and µn = 1 (13) where µ n and ν are the Lagrange multipliers for multimodal terms and volume term respectively. The Lagrange multipliers are determined by maximising the complimentary Lagriangian, L C, subject to non-negativity of the Lagrange multipliers, µ n 0. The resulting optimisation problem can therefore be formulated as, [ N ( ) ] v max min µ n r n + ν 1 (14) µ n,ν V i,w i,h i v 0 n=1 subject to the afore mentioned constraints. The goal is now to develop a separable approximation of equation (14), which can be solved locally for each design region. IJsselmuiden et al. [5] demonstrated that the inverse buckling factor can be approximated as, r N [ Ψ m i : A i + Φ b i : D i] 1 i=1 where Ψ m i, and Φ b i are the sensitivity matrices with respect to in-plane stiffness and out of plane compliance respectively, as is outline in [5]. The approximation is not strictly convex in terms of the in-plane stiffness and therefore these terms are expanded linearly (15)

5 in terms of the design variables. Neglecting constant terms and those that sum to zero the linear expansion for iteration k + 1 can can be expressed as, Ψ m i : A i Ψ m i : (Âi h (k) i ) + Â(k) i h i where ˆ represents the part of the stiffness term which is independent of thickness. Since both objective function and constraints are separable, the buckling load can maximised by solving the following local optimisation problem at each design point, (k) min Ψm i : Âih i + Φ b 1 12 ( ) i : ˆD i + Ψm V i,w i,h i h 3 i : Â(k) i + ν h i (17) i where the sensitivities of the local optimisation account for all the active buckling modes, Φ b i = N n=1 µ n Φ b i,n and Ψm i = (16) N µ n Ψ m i,n (18) Even though the proposed approximations are convex, the buckling approximation is not strictly convex [5]. Due to this lack of strict convexity, convergence problems are encountered when maximising the buckling load factor. The presence of multimodal designs also further compounds the convergence difficulties. The situation can be remedied by regularising the expression by introducing an additional term which is strictly convex with respect to the design variables. The proximal point algorithm, following Rockafellar [11], is an effective method of ensuring convergence while retaining a separable approximation. The local optimisation problem is therefore expressed as, n=1 ( min Ψ m i V i,w i,h i : Âih (k) i + Φ b i : 1 12 ( ) ˆD i + Ψm h 3 i : Â(k) i + ν h i + η ) i 2 C i (19) where η is a scaling factor which is free to be defined. The convex term C is defined as, C i = 4 [ V ij Ṽij 2 + W ij W ] ij 2 j=1 Note that the contribution of this term tends to zero as the solution converges. The value chosen for η is constant and in essence represents a move limit for the design variables, hence influencing the speed of convergence. Smaller values of η result in quicker convergence, however as η tends to zero the local approximation may no longer be strictly convex. (20) PRELIMINARY RESULTS The purpose of this section is to illustrate the benefit of using variable stiffness design with lamination parameters, as well as to study the mechanisms resulting in improved buckling loads. An example problem, previously studied by Olmedo and Gürdal [10], is investigated. The test case consists of a simply supported rectangular plate subject to axial compression introduced via uniform edge displacement, i.e. the edges remain straight. In

6 this case a square configuration with side lengths of 15 inch and an average thickness of 0.06 inch is investigated based on the following orthotropic material properties: E 11 = Msi, E 22 = 1.49 Msi, G 12 = 1.04 Msi, ν 12 = The plate is discritised into a selected number of equally sized elements and analysed using a finite element routine programmed in Matlab TM. Due to high computational times, the presented results are for a relatively course 100 element mesh, resulting in 605 design variables (5 per node). Results are normalised with respect to the quasi-isotropic buckling load, which is found to be 519 lb. The developed optimisation routine is used to compute the optimal thickness and stiffness distributions considering different lower bounds on laminate thickness. The first four buckling modes, ratio of maximum to minimum thickness, and maximum thickness are listed in Table 1. Results for constant stiffness and variable stiffness laminates with uniform thickness are also presented for comparison. Table 1: Normalised buckling loads for a range of plate designs. Design ˆλ1cr ˆλ2cr ˆλ3cr ˆλ4cr h max /h min h max Qausi Isotropic Constant Stiffness Variable Stiffness Variable stiffness with h ave /h min : Results demonstrate that tailoring both stiffness and thickness dramatically improves the buckling performance of a plate. Improvements in the order of 200 to 440% are achieved, compared to 150% when only stiffness properties are tailored. Several trends are visible in the results. Firstly the mode spacing tends to decrease as the bound on minimum thickness decreases. Hence by increasing the design freedom, more modes become active during the optimisation. The thickness distribution also tends to force the modes to localise towards the centre of the plate, as can be seen by inspecting Figure 1 and Figure 2. Secondly, the maximum thickness does not increase beyond inch, even when the lower bound on thickness is decreased further. The ratio of maximum to minimum thickness however does continue to increase, indicating that the lower bound on thickness is active. Therefore when h min < h ave /2, the improved buckling loads are no longer as a result of increasing edge thickness. This can be seen by inspecting the thickness distribution of the different plates in Figure 3. As h min decreases the area of minimum

7 (a) Mode 1 (b) Mode 2 (c) Mode 3 (d) Mode 4 Figure 1: First four buckling modes for h min = 0.9h ave (a) Mode 1 (b) Mode 2 (c) Mode 3 (d) Mode 4 Figure 2: First four buckling modes for h min = 0.3h ave thickness at the centre of the plate also decreases. Thickening towards the edge of the plate indicates that load redistribution is the primary mechanism resulting in improved buckling loads. The same conclusion was found in ref. [5], where only stiffness properties were tailored. Clearly combining both thickness and stiffness tailoring compounds their individual effects yielding the largest possible improvement. CONCLUSIONS A separable, conservative approximation scheme was developed allowing lamination parameter and thickness distributions to be computed for maximum buckling loads. Preliminary results confirmed that load redistribution was the driving mechanism behind buckling load improvements. Combining both thickness tailoring and stiffness variations via steering yielded substantially higher buckling loads than if only thickness or stiffness variations was considered separately. Improvements in the order of 440 % with respect to a qausi-isotropic design were demonstrated. The optimal thickness distribution tends to localise the buckling modes towards the centre of the plate. In future work the optimisation routine will be extended to incorporate automatic scaling of the proximal term, allowing smooth, efficient convergence of the optimisation. This will allow a larger set of design problems to be investigated in order to identify more general mechanisms responsible for design improvements. Additionally the formulation will be extended to incorporate shell type structures, for which in-plane and bending coupling is present, and hence designs will be more sensitive to thickness variations. ACKNOWLEDGEMENTS This work is supported by the AUTOW project, part of the European Union Sixth Framework Programme.

8 Ly Ly Lx (a) h min = 0.9h ave Lx (b) h min = 0.7h ave Ly Ly Lx (c) h min = 0.5h ave Lx (d) h min = 0.3h ave 0.4 Figure 3: Normalised optimal thickness distribution, h/h ave, for different minimum thickness bound. APPENDIX: MATERIAL INVARIANTS Lamination parameters allow laminate stiffness properties to be defined as a linear combination of laminate invariant matrices (see equations 9 and 10): U 1 U 4 0 U U 2 /2 Γ 0 = U 4 U 1 0, Γ 1 = 0 U 2 0 Γ 2 = 0 0 U 2 /2, 0 0 U U 2 /2 U 2 /2 0 U 3 U U 3 Γ 3 = U 3 U 3 0, Γ 4 = 0 0 U U 3 U 3 U 3 0 where the laminate invariants, U i for i = , are defined in terms of lamina reduced

9 stiffnesses by: [6] U 1 = (3Q Q Q Q 66 )/8, U 2 = (Q 11 Q 22 )/2, U 3 = (Q 11 + Q 22 2Q 12 4Q 66 )/8, U 4 = (Q 11 + Q Q 12 4Q 66 )/8, U 5 = (Q 11 + Q 22 2Q Q 66 )/8. REFERENCES [1] Mostafa M. Abdalla, Shahriar Setoodeh, and Zafer Gürdal. Design of variable stiffness composite panels for maximum fundamental frequency using lamination parameters. Composite Structures, 81(2): , [2] Sherrill B. Biggers and Sundar Srinivasan. ompression buckling response of tailored rectangular composite plates. AIAA Journal, 31(3): , [3] Claude Fleury and Lucien A. Schmit Jr. Dual methods and approximation concepts in structural synthesis. NASA Contractor Reports, (3226), [4] V. B. Hammer, M. P. Bendsøe, R. Lipton, and P. Pedersen. Parametrization in laminate deign for optimal compliance. International Journal of Solids and Structures, 34(4): , [5] Samuel T. IJsselmuiden, Mostafa M. Abdalla, Shahriar Setoodeh, and Zafer Gürdal. Design of variable stiffness panels for maximum buckling load using lamination parameters. In 49th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, number AIAA , Schaumburg, IL, April AIAA. [6] R. M. Jones. Mechanics of Composite Materials. Taylor & Francis, Inc., 2nd edition, [7] Makarand G. Joshi and Sherrill B. Biggers. Thickness optimization for maximum buckling loads in composite laminated plates. Composites Part B: Engineering, 27(2): , [8] Christos Kassapoglou. Composite plates with two concentric layups under compression. Composites Part A: Applied Science and Manufacturing, 39(1): , [9] N. Olhoff. Multicriterion structural optimization via bound formulation and mathematical programming. Structural Optimization, 1:11 17, [10] R. Olmedo and Zafer Gürdal. Buckling response of laminates with spatially varying fiber orientations. In AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 34th and AIAA/ASME Adaptive Structures Forum, pages , 1993.

10 [11] R.T. Rockafellar. Monotone operators and the proximal point algorithm. SIAM Journal Control and Optimization, 14(5): , August [12] Shahriar Setoodeh, Mostafa M. Abdalla, and Zafer Gürdal. Approximate feasible regions for lamination parameters. In 11th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, Portsmouth, Virginia, [13] A. P. Seyranian, E. Lund, and N. Olhoff. Multiple eigenvalues in structural optimization problems. Structural Optimization, 8(4): , [14] S. W. Tsai and H. T. Hahn. Introduction of Composite Materials. Technomic Publishing Company, Lancaster, PA, [15] S. W. Tsai and N. J. Pagano. Invariant properties of composite materials. Composite Material Workshop, pages , 1968.

Design and optimization of a variable stiffness composite laminate

th World Congress on Structural and Multidisciplinary Optimisation 07 th - th, June 05, Sydney Australia Design and optimization of a variable stiffness composite laminate Yan Zhang, Fenfen Xiong Qian